ABSTRACT Identification of high-quality chemical probes, molecules with high specificity and selectivity against macromolecules, is of critical interest to drug discovery. Although millions of compounds have been screened against thousands of protein targets, small-molecule probes are currently available for only 4% of the human proteome. Thus, more efficient approaches are required to accelerate the development of novel, target-specific probes. In 2019, a new bold initiative called “Target 2035” was launched with the goal of “creating […] chemical probes, and/or functional antibodies for the entire proteome” by 2035. In support of this ambitious initiative, we propose to develop and test a novel integrative AI-driven methodology for rapid chemical probe discovery against any target protein. Here, we will build an integrative workflow where the unique XChem database of experimental crystallographic information describing the pose and nature of chemical fragments binding to the target protein will be used in several innovative computational approaches to predict the structure of organic molecules with high affinity towards specific targets. The candidate molecules will be experimentally validated and then optimized, using computational algorithms, into lead molecules to seed chemical probe development. The proposed project is structured around three following interrelated keystones: (i) Develop a novel method for ligand-binding hot-spot identification and discovery of novel chemical probe candidates; (ii) Develop novel fragment-based integrative computational approach for accelerated de novo design of chemical probes; (iii) Consensus prediction of target-specific ligands, synthesis, and experimental validation of computational hits. More specifically, we will develop a hybrid method to predict structures of high-affinity ligands for proteins for which XChem fragment screens have been completed. These approaches will be used for screening of ultra- large (>10 billion) chemical libraries to identify putative high affinity ligands within crystallographically determined pockets. Then, we will develop and employ an approach using graph convolutional neural networks for de novo design of a library of strong binders that will be evaluated to select the best candidates for chemical optimization. Finally, we will combine traditional structure-based and novel approaches, developed in this project to select consensus hit compounds against three target proteins: transcription factor brachyury, hydrolase NUDT5, and bromodomain BAZ2B. Iterative design guided by the computational algorithms, synthesis, and testing will progressively optimize molecules to micromolar leads to chemical probes for the target proteins. Completion of the proposed aims will deliver a robust integrative workflow to identify leads for chemical probes against diverse target proteins. We expect that our AI-based computational approach to convert crystallographically-determined chemical fragments into lead compounds coupled with the experimental validation of computational algorithms will accelerate the discovery of new chemical probes, expand the druggable proteome, and support future drug discovery studies

PROJECT NARRATIVE This project is devoted to the development of a novel integrative, structure-based, artificial intelligence driven methodology for rapid chemical probe discovery against any target protein. The candidate molecules will be experimentally validated for three selected targets, transcription factor brachyury, hydrolase NUDT5, and bromodomain BAZ2B, and then optimized, using computational algorithms, into lead molecules to seed chemical probe development. We expect that our AI-based computational approach to convert crystallographically- determined chemical fragments into lead compounds coupled with the experimental validation of computational algorithms will accelerate the discovery of new chemical probes, expand the coverage of druggable proteome, and support future drug discovery studies.